Re: [rdiff-backup-users] Moving the backup-location to another machine

[Chris Wilson]

Hi Ron,

On Sun, 12 Jan 2014, Ron Leach wrote:

> > It seems that someone reduced the number of reserved blocks on your old 
> > system to zero.
>
> That was set by the Debian installer, in fact, but it was xfs, though, 
> not ext4.  This new lv is ext4.

Ok, perhaps XFS does not have such a feature. ext* does, and this explains 
why you have some blocks missing/not usable.

> > It's an extremely bad idea in my view to run an ext4 filesystem over 
> > 80% full, as it results in increasing fragmentation over time and there 
> > is no defragmentation tool available.
>
> Very interesting point.  We keep the backup increments indefinitely so, 
> other than rdiff-backup housekeeping files, the bulk of the filesystem 
> stays in existence.  No deletion, in practice.

That doesn't matter. New files, especially large ones, will be forced to 
squeeze into the increasingly small gaps left in each inode group, and 
thus be fragmented from birth.

> I'm not sure if the reverse-diffs change on each backup run

No, new ones are created and the old full image files are deleted whenever 
a file changes. (I think).

> Chris, that was an extremely helpful post; it has pointed me towards a 
> more critical appraising of suitable filesystems.

Thanks, I'm glad. But to be honest, I don't think there's a filesystem 
that can avoid fragmentation when 99% full except for a pure 
log-structured one, and you'll have other problems there. (Can't delete 
files when the filesystem is full; may not be able to resize either.)

Nor have I ever seen a good reason to avoid ext* for any task except:

* huge numbers of very small files (for which the cure at the time was 
reiserfs, now deprecated I think, so no excuse any more);

* needing live snapshots (which only BSD UFS2, ZFS and the experimental 
BTRFS support, afaik);

* needing write-once integrity (for which you need a log-structured 
filesystem such as nilfs);

* needing to write directly to NAND flash (for which you need yaffs or 
ubifs);

* needing live mirroring to another filesystem, dynamic volume pools 
within a filesystem or in-filesystem RAID (for which ZFS is the only 
option that I know of).

Cheers, Chris.
--
_____ __     _
\  __/ / ,__(_)_  | Chris Wilson <address@hidden> Cambs UK |
/ (_/ ,\/ _/ /_ \ | Security/C/C++/Java/Ruby/Perl/SQL Developer |
\__/_/_/_//_/___/ | We are GNU : free your mind & your software |